Convective storms are one of the most hazardous weather phenomena in aviation due to their unpredictability and the accompanying strong winds, turbulence, icing and lightning. Information on deep convection clouds in meteorological reports are therefore of great importance to pilots. These reports have to be issued by meteorological services regularly according to international conventions, so they need to be able to issue automated reports in case of airport closure or absence of meteorologists observers. In Slovenia automated meteorological reports are issued using a number of sensor measurements, however, no object identification methods using radar measurements have been tested for this purpose so far.
In this master's thesis I have developed a method for detection and categorisation of deep convection clouds using measurements of radar reflectivity. I considered the period from May to September for years 2018 and 2019. FiT algorithm (Forward in Time) was used for object identification based on data prepared using maximum reflectivity ($Z_\text{max}$) and vertically integrated liquid ($VIL$). Meteorological reports for Ljubljana airport were used for verification. Success of each vertical aggregation method was evaluated with the use of contingency table through calculating success indexes such as CSI (Critical Success Index). Both $Z_\text{max}$ and $VIL$ categorised storm clouds (Cb) with CSI value of 0,558. Monthly analysis using optimal parameters showed large variability of index bias and both method options were most successful in June. In order to improve the results, I found optimal parameters of each month separately. This approach has proven to be slightly more successful, but more data should be analysed for results to be representative. Only $Z_\text{max}$ was successful in categorisation of towering cumulus clouds (TCu) and reached a CSI of 0,3855. Poorer performance was achieved due to a bigger number of false alarms. Using optimal parameters for categorisation of both cloud types the $Z_\text{max}$ method option showed a CSI of 0,5030. In the end, performance of optimal settings was tested on an independent set of data, the 2020 convective season, and a CSI value of 0,4320 was reached.
|